Learning subtree pattern importance for Weisfeiler-Lehman based graph kernels

Graph is an usual representation of relational data, which are ubiquitous in many domains such as molecules, biological and social networks. A popular approach to learning with graph structured data make use kernels, measure the similarity between graphs plugged into a kernel machine support vector machine. Weisfeiler-Lehman (WL) based employ WL labeling scheme extract subtree patterns perform node embedding, demonstrated achieve great performance while being efficiently computable. However, one main drawbacks general decoupling construction process. For molecular graphs, kernels subtree, on substructures consider all available having same importance, might not be suitable practice. In this paper, we propose method learn weights framework WWL state art for classification task (Togninalli et al., in: Advances Neural Information Processing Systems, pp. 6439–6449, 2019). To overcome computational issue large scale sets, present efficient algorithm also derive generalization gap bound show its convergence. Finally, through experiments synthetic real-world demonstrate effectiveness our proposed patterns.